12-04, 19:00–19:30 (UTC), General Track
Julia is a high-performance language for technical computing that offers advantages like type stability, just-in-time compilation, and extensive parallel computing support. Its Machine Learning ecosystem, although having fewer options, is functional and includes packages like DataFrames.jl, Flux.jl, MLJ.jl, and SciML for various ML tasks. Additional tools cover data visualization, R compatibility, and specific ML applications. The ecosystem is comprehensive and can meet many ML researcher/professional needs. This talk provides an overview of the ecosystem, discussing both its strengths and potential areas for improvement.
Julia is a high-level, high-performance dynamic language for technical computing. While Python is more general-purpose and it clearly has a larger user community, Julia has many interesting advantages, like type stability, just-in-time compilation, and extensive support for parallel and distributed computing, including an excellent usage of GPU and TPU. These features, in combination with its ability to easily interact with other languages such as Python or R, are very useful for Machine Learning.
The Julia ecosystem for Machine Learning (ML)) is, although limited in options, very completed in functionality. Some noteworthy Julia packages for ML are:
- As Alternative to Pandas, [[https://dataframes.juliadata.org/stable/][DataFrames.jl]]: It provides data structures and operations for handling tabular data.
- For Deep Learning, there are several, the most popular is [[https://fluxml.ai/Flux.jl/stable/][Flux.jl]] with its own little [[https://fluxml.ai/Flux.jl/stable/ecosystem/][ecosystem]].
- Similar to sickit-learn: [[https://juliaai.github.io/MLJ.jl/stable/][MLJ.jl]]: A framework for supervised and unsupervised learning that supports model selection, tuning, and ensembling. It also provides a standardized API for other ML libraries in Julia like Flux, [[https://github.com/JuliaAI/DecisionTree.jl][DecisionTree.jl]] .jl, or the library scikit-learn in Python, through Scikit-learn.jl.
- The [[https://docs.sciml.ai/Overview/stable/][SciML organization]] provides a suite of packages for scientific machine learning, including differential equations, optimization, and uncertainty quantification.
- For R users, there is a group of packages grouped [[https://github.com/TidierOrg/Tidier.jl][Tidier.jl]], that replicates as much as possible well-known libraries interface in R, like ggplots (TidierPlots) or dplyr and tidyr (TidierData).
- Visualization packages, like [[https://docs.juliaplots.org/latest/][JuliaPlots.jl]], or [[https://docs.makie.org/stable/][Makie.jl]] (including other more higher-level as [[https://aog.makie.org/stable/][AlgebraOfGraphics.jl]]).
- Other Specific ML libraries, like the Symbolic Regression library [[https://github.com/MilesCranmer/SymbolicRegression.jl][SymbolicRegressions.jl]] accessible in Python as [[https://github.com/MilesCranmer/PySR][PySR]] library.
As can be seen, it is quite a broad ecosystem that can fulfill many needs of a Machine Learning researcher/professional. Since it is very broad, I will focus more on the components that the public themselves consider to be of most interest to them.
This talk give first a general snapshot of the different tools, and them a more critical point of view of how ML in Julia is currently, which are the strong features, and also their disadvantages, in a critical way.
This talk is addressed to people with knowledge in Machine Learning using Python, and give them the current state of Julia in ML, in which parts they could incorporate them, and how they could enrichment their workflow using it, and also in which parts they could improve to the ML Julia community, if they are interested.
No previous knowledge expected
As many people, I have several roles:
As a computer science that love programming, I am interested in programming in different languages (like C++, Java, ....), and I particularly love Python, and another interesting growing programming languages (like Rust or Julia). I define myself also as a Linux user, this is the only OS in my computers for more than 15 years. Also, I am a believer in Free Software (actually, I was for years the secretary of Free Software Office at the University of Cadiz). I usually give talks in PyData in my country, and I also I have participated in several JuliaCon conferences.
Since a professional side, I am Assistant Professor at the University of Granada, Spain, in Computer Science. I research in Artificial Intelligence using Metaheuristics for optimization (I have won two international competitions) and also in Neuroevolution, combining Metaheuristics with Deep Learning, with an index-h of 31, and be member of the 2% most influencer researcher in Stanford's list. I have directed two thesis. I have participated in several research projects involving Machine Learning. i am currently co-leading a General Purpose Artificial Intelligence project, a €120K Knowledge Generation Project, funded by the Ministry of Science, Innovation and Universities of Spain.